Spontaneous recovery of version by canonical correlation analysis network
Pei Ling Lai · 2003
We review a method of Lai and Fyfe (1999) for performing canonical correlation analysis with artificial neural networks. We demonstrate its capability on a simple artificial data set and then on a real data set where the results are compared with those achieved with standard statistical tools. We extend the method by implementing a very precise set of constraints which allow a locally linear version of the CCA network, using a lateral matrix of connections to order the linear correlations. We demonstrate the network's capabilities on artificial data and on the stone's data set.